针对复杂工况下水泵机电运行参数趋势预测的问题，建立基于多任务学习(multi-task learning，MTL)和注意力机制(attention mechanism，AM)的水泵机电运行参数趋势预测模型。充分利用历史工况数据，使用多任务学习分析方法，寻找历史工况数据的共同特征；在预测新工况数据变化趋势时，引入注意力机制动态分配共同特征映射时的权重系数，突出关键共同特征，提升模型的预测精度；根据模型监测统计量阈限分析，建立机组运行监测多级预警模型，优化运维管理策略。以某泵站机组实际运行工况数据进行测试并与不同模型计算结果进行对比分析，结果表明：与传统单任务学习和静态共同特征映射权重的模型相比，基于多任务学习和注意力机制的模型，其统计量T2和Q均未超过95%和99%的控制限，表明该预测模型具有很好的稳定性和准确性。

The safe and stable operation of the pumping station system is of great significance for ensuring supply for domestic water agricultural irrigation, and industrial water. Therefore, real-time monitoring of pump station operating parameters and establishing predictive models for fault diagnosis and intelligent alarm of unit equipment have significant application value. The data-driven method for fault diagnosis is currently a hot topic in the research of pump station equipment status monitoring. However, there are problems such as insufficient data samples, difficulty in feature extraction, and insufficient generalization ability in practical application. Addressing the challenge of predicting the trends in operating parameters of water pump units under complex working conditions, a prediction model for operating parameters of water pump units was proposed based on multi-task learning method and attention mechanism. Firstly, the historical working condition data was fully utilized, and a multi-task learning model was established to find the common characteristics of the historical working condition data on the basis of traditional principal component analysis methods. Secondly, an attention mechanism was introduced to dynamically allocate weight coefficients for common feature mapping when predicting the trend of parameter changes under new operating conditions, highlighting key common features and improving the accuracy of the prediction. Based on the actual operating data of a pumping station hub unit, the performance of the model was tested. By monitoring the statistical parameters T2 and Q, which reflecting the stability and accurately of the model, results showed that the prediction model proposed has good stability and prediction accuracy under 98% and 95% control thresholds. On this basis, a multi-level equipment operation monitoring and alarm model was also preliminarily established. The alarm level is divided into three levels: yellow, orange, and red. Management personnel can take different disposal measures based on the alarm level, such as paying attention to observation, strengthening monitoring and inspection frequency, adjusting equipment operation, and even shutting down for maintenance, to avoid accidents and ensure the safety of equipment and personnel. By comparing the predictive performance of the proposed method with single task learning and the model without attention mechanism, it can be concluded that: Compared with traditional static PCA model prediction methods, the multi-task learning model can fully utilize the common features of historical data to predict changes in unit parameters, fully consider the correlation between different tasks, and improve the robustness of the prediction model. The introduction of attention mechanism enables the model to dynamically adjust the mapping weights based on the characteristics of unit operating parameters in new time periods, further improving the stability of the model and the accuracy of prediction. The results have important application value for safety monitoring and intelligent warning of pump unit operation in pumping stations.